sourceImage xlink:href=”http:laits.gmu.eduWCSInputCovaerageMODI
S.hdf” Format HDF-EOS Format
sourceImage trainedParametersID
http:laits.gmu.eduWCSInputparametersForMODIS.hdf trainedParametersID
classifiedCoverageFormat HDF-EOS classifiedCoverageFormat GetClassification
9.4.2 GetClassification response
The response to a request for classifying an image coverage to a thematic coverage the server could be, if no exception occurs:
?xml version=1.0 encoding=UTF-8? Classification xmlns=http:www.opengis.netwics
xmlns:xlink=http:www.w3.org1999xlink xmlns:xsi=http:www.w3.org2001XMLSchema-instance
xsi:schemaLocation=http:www.opengis.netwics ..wics.xsd service=WICS version=1.0.0
classifiedCoverage xlink:href=”http:laits.gmu.eduWCSInputCovaerageClassifiedMODIS.hdf”
Format HDF-EOS Format classifiedCoverage
Classification
10 TrainClassifier
10.1 Introduction
The TrainClassifier operation allows WICS clients to train a supervised classifier supported by WICS with client supplied training data. Training data include two separate
data sets which shall be stored in two separate data files. One data set contains the image data on which training is to be performed. The image data is usually three-dimensional
data array, although in very limited occasions two-dimensional might also be allowed. The other data set contains a two-dimensional data array whose values are class labels.
The two data sets must have the same numbers of rows and columns and must be geographically co-registered. WICS clients must encode the training data based on the
encoding format supported by the server. The training data sets shall be provided as two URLs referenced to the two data set files.
10.2 Operation request
10.2.1 TrainClassifier request parameters
A request to perform the TrainClassifier operation shall include the parameters listed and defined in Table 12. This table also specifies the UML model data type, source of values,
and multiplicity of each listed parameter, plus the meaning to servers when each optional parameter is not included in the operation request. Parameters common to the other
32 © Open Geospatial Consortiium 2005
operations which are listed in Table 2, i.e., Service, Request, and Version, are not re- listed in the current table.
Table 12 — Parameters in TrainClassifier operation request
Name Definition
Data type and value Multiplicity and use
TrainingImage Image coverage data
on which training shall be performed
One URL or one operation request which returns an image coverage
One Mandatory TainingLableI
mage Two-dimensional
grid coverage whose values are
known class labels. One URL or one operation request
which returns an image coverage One Mandatory
TrainingImage Format
Format of the input training grid image
coverage Character String type, not empty
Must be one of those listed under the sourceImageFormats of the
classifier’s supportedFormats One Mandatory
TrainingLableI mageFormat
Format of the training label image
Character String type, not empty Must be one of those listed under
the sourceImageFormats of the classifier’s supportedFormats
One or zero optional
Assumed to the same as the
TrainingImageForm at if not present
ClassifierPara meters
The same as defined in Table 7,
subclause 9.2.1 of this document
The same as defined in Table 7, subclause 9.2.1 of this document
One Mandatory
The response of the TrainClassifier, trainedParameterID, shall be a UUID of the trained parameters data file generated from the training process. This data file shall be kept at the
server side and the encode format is internal to the classifier. Thus, the requesting client does not need to know this internal encoding format. The server may announce an
expiration date and time of this UUID if the server will not keep this file permanently. The client can use this trained parameter data in subsequent GetClassification request
through referencing to the UUID before the expiration date and time. Alternatively, the client might also store this data file in client side or any other locations and generate a
UUID for the stored file. In this way, the client might be able to use this file after the expiration date and time. However, it should be noted it is very likely that a client may
not be able to correctly keep track of the trained parameter data files after store them in client side or remote locations, especially after multiple TrainClassifier operations are
sent to multiple servers. Thus, in order to make sure that the trained data parameters submitted to it is correct, a WICS server may not allow clients to actually get this file.
10.2.2 TrainClassifier request KVP encoding optional
